Let us start from the premise that every person's desires are equally important. Yet everybody knows the flavors of democracy implemented in the last few centuries can produce terrible outcomes. For that, you don't need to agree that the election of Bush was a less than optimal outcome --something that the majority of the American public now seems to believe. It suffices to believe that the election of the Nazi party to become the largest party in the Reichstag (German Parliament) was a bad idea. The harder question is, is there any better system?

To address that, we need to understand the shortcomings of democracy. Perhaps the biggest shortcoming is that people who are ignorant about the consequences of voting for any given candidate, and who thus vote against their own interests, get to vote too, and their vote counts just as much as that of those who know exactly what they are getting into.

A better system needs to ferret out the ignorance in the system. Below, I propose two alternatives to our current system that may do the job.

The first proposal squeezes the ignorance out of the system by administering a set of multiple-choice questions to each voter. The questions would be designed to test the knowledge of the voter about the issue being voted on. If the vote is for a candidate, they might be specific questions about the candidates' proposals for government, about their prior voting record, or about their track record. The vote of each voter would then be weighed by their score in the test, e.g. by the proportion of correct answers. Note that this is not an intelligence test, or a general knowledge test, simply a test of the factual issues at hand in that particular election. Clearly this would weigh more heavily the votes of those who know what they are really voting about.

To produce the list of questions in the test, the system used for jury selection would be used: a body would produce candidate questions (the body could be an ad-hoc commission, a committee of Congress, even proposals by ordinary citizens), and each major party (as defined by % of votes in the previous election) would have a veto right on any question, with a maximum number of vetos reserved for each party to ensure vetos were used judiciously.

The second proposal is less certain to work. It is based on the principle that people have better local knowledge than global knowledge. In other words, they know the people they interact with better than those they don't. In this system, called the local representative system, each person casts two votes: one for a candidate in the election, and another for his representative. The representative cannot be a candidate, but needs to be someone the voter knows personally. Although this seems hard to enforce today, digital social networks are becoming pervasive enough that there will be soon enough digital information to find out who the N individuals most connected to any person are. The network of votes for representatives is used to compute the PageRank of each representative. The vote for a candidate by each representative is then weighed by the representative's PageRank, which is a measure of the representative's representativeness. The idea here is that, while people may be ignorant about candidates, they will on average be less ignorant about the people they know, and will be able to make a better decision about who among them they would trust their vote with.

The difficulty with this second proposal is that it will only produce results that are different from conventional 20the century democracy if a meaningful number of representatives vote differently from the people who voted for them. Which means the system can be "gamed" by people voting for people who they know would vote for the same candidate as they do. This assumes that everybody thinks they know who they want as president. But a non-negligible fraction of the electorate might be aware of their lack of knowledge, and be prepared to proxy their vote to someone they trust. One might ask why this cannot be achieved by people asking for the proxy's advice before voting. But not every proxy may be prepared to dole out advice, and more importantly, PageRank captures the trust conveyed by the population on each voter in a way that isolated pairwise interactions, like advice, cannot.

Although the voting system has complications that are not present in the WWW, in some sense to believe that this would work better than our current democracy is no more revolutionary than believing that Google's PageRank works better at finding good webpages than a simple link count.

Obviously, both of these proposals would be easier to implement with electronic voting.

How would one test whether the proposed systems work better than our current flavor of democracy? One way is to implement them in some scale, say as an application on a social network, compute the candidates that the system would elect for a set of elections, and then compare the approval ratings at the end of office for candidates elected who would also have been elected by the new system, with ratings for candidates elected who would also have been elected through ordinary democracy by the same subset of voters. One place to start for the local voting principle is to test whether approval ratings are higher at end of office for very small electorates (e.g. towns with less than 500 people) than in large ones. But there might be all sorts of reasons why smaller towns might be more or less content, so that's not a great measurement.

Any charitable foundation willing to put the money to put this to a test? Surely improving on our centuries-old system of elections is worth an attempt...

--Alex Backer

The above ideas are excerpted from a conversation between Alex Backer and Eric Slimko on 7/31/2008.